tarcs: a topology change aware-based routing protocol

electronics

Article

TARCS: A Topology Change Aware-Based RoutingProtocol Choosing Scheme of FANETs

Jie Hong 1,2,* and Dehai Zhang 1

1 Key Laboratory of Microwave Remote Sensing, National Space Science Center, Chinese Academy ofSciences, Beijing 100190, China; [email protected]

2 University of Chinese Academy of Sciences, Beijing 100094, China* Correspondence: [email protected]; Tel.: +86-131-4603-6664

Received: 21 December 2018; Accepted: 21 February 2019; Published: 2 March 2019��

Abstract: The rapid change of topology is one of the most important factors affecting the performanceof the routing protocols of flying ad hoc networks (FANETs). A routing scheme suitable for highlydynamic mobile ad hoc networks is proposed for the rapid change of topology in complex scenarios.In the scheme moving nodes sense changes of the surrounding network topology periodically, andthe current mobile scenario is confirmed according to the perceived result. Furthermore, a suitablerouting protocol is selected for maintaining network performances at a high level. The concernedperformance metrics are packet delivery ratio, network throughput, average end-to-end delay andaverage jitter. The experiments combine the random waypoint model, the reference point groupmobility model and the pursue model to a chain scenario, and simulate the large changes of thenetwork topology. Results show that an appropriate routing scheme can adapt to rapid changes innetwork topology and effectively improve network performance.

Keywords: flying ad hoc network (FANET); mobile ad hoc network (MANET); highly dynamic;periodical; topology change awareness; routing protocol

1. Introduction

Mobility is one of the most prominent features of mobile ad hoc networks (MANETs) and one ofthe important factors affecting network performance. Being a typical subset of MANETs, the nodemobility in flying ad hoc networks (FANETs) [1–3] is stronger and severely impacts the networkperformance (Figure 1).

Electronics 2018, 7, x; doi: FOR PEER REVIEW www.mdpi.com/journal/electronics

Article

TARCS: A Topology Change Aware-Based Routing Protocol Choosing Scheme of FANETs Jie Hong 1,2,* and Dehai Zhang 1

1 Key Laboratory of Microwave Remote Sensing, National Space Science Center, Chinese Academy of Sciences, Beijing 100190, China; [email protected]

2 University of Chinese Academy of Sciences, Beijing 100094, China * Correspondence: [email protected]; Tel.: +86-131-4603-6664

Received: 21 December 2018; Accepted: 21 February 2019; Published: date

Abstract: The rapid change of topology is one of the most important factors affecting the performance of the routing protocols of flying ad hoc networks (FANETs). A routing scheme suitable for highly dynamic mobile ad hoc networks is proposed for the rapid change of topology in complex scenarios. In the scheme moving nodes sense changes of the surrounding network topology periodically, and the current mobile scenario is confirmed according to the perceived result. Furthermore, a suitable routing protocol is selected for maintaining network performances at a high level. The concerned performance metrics are packet delivery ratio, network throughput, average end-to-end delay and average jitter. The experiments combine the random waypoint model, the reference point group mobility model and the pursue model to a chain scenario, and simulate the large changes of the network topology. Results show that an appropriate routing scheme can adapt to rapid changes in network topology and effectively improve network performance.

Keywords: flying ad hoc network (FANET); mobile ad hoc network (MANET); highly dynamic; periodical; topology change awareness; routing protocol

1. Introduction

Mobility is one of the most prominent features of mobile ad hoc networks (MANETs) and one of the important factors affecting network performance. Being a typical subset of MANETs, the node mobility in flying ad hoc networks (FANETs) [1–3] is stronger and severely impacts the network performance (Figure 1).

MANET

VANET

FANET

Figure 1. MANET, VANET and FANET. Figure 1. MANET, VANET and FANET.

FANET has some unique characteristics compared to MANET and VANET (Table 1). In FANETs,high-speed moving nodes usually need to complete multiple tasks, such as search/exploration,

Electronics 2019, 8, 274; doi:10.3390/electronics8030274 www.mdpi.com/journal/electronics

http://www.mdpi.com/journal/electronics

http://www.mdpi.com

https://orcid.org/0000-0003-2739-9210

http://www.mdpi.com/2079-9292/8/3/274?type=check_update&version=1

http://dx.doi.org/10.3390/electronics8030274

http://www.mdpi.com/journal/electronics

Electronics 2019, 8, 274 2 of 19

reconnaissance/patrol, or target tracking. Different task scenarios could have corresponding nodemobility models [1]. When the task changes, the node mobility mode (such as node motion speed,direction, distance, etc.) changes accordingly, which will cause the network topology to change rapidlyand seriously affect the network performance.

Table 1. Comparison of MANET, VANET and FANET.

Characteristics MANET VANET

FANET

Small UAVs Large UAVs

FW * RW * FW * RW *

Node Speed Low Medium Medium Medium High HighDensity High High Low

Power Consumption Low High Med Med High HighNetwork

Connectivity High Medium Low

QoS Medium Depends on the applicationsMobility Patterns 2D 2D 3DMobility Degree Low Medium Low-Med Low-Med Med-High Med-High

Altitude Close to ground Close to ground Low Low High HighStatic Pause Yes Yes No Yes No Yes

Topology Variation Occasionally Frequently Very frequently Very frequentlyscalability Medium High Low

Typical MobilityModels regular Restricted

through roads Depends on the applications and need to pre-determined

* FW: fixed-wing, * RW: rotary wing.

The main research object of this paper is highly dynamic FANETs. Specifically, it corresponds to theFANETs listed in Table 1 in which the nodes move fast and the network topology changes drastically.

The characteristics of the FANETs discussed in this paper are: small and medium-sized networks,high mobility nodes far from the ground at high altitudes, sufficient node energy and frequent networktopology changes.

In practical applications, FANETs encounter different application scenarios when nodes completecomplex tasks. Nodes move in different ways in different scenarios, which causes large changes inthe network topology. Traditionally, nodes cannot perceive this change, nor can they adapt quicklyto it, so the network performance will deteriorate dramatically. The main purpose of this paper is tofind a method by which nodes in a network can perceive changes of topology and adjust the routingstrategy appropriately when the topology changes greatly, so that the network performance couldbe maintained at a high level. The most concerned network performance is the packet delivery ratio,followed by the average end-to-end delay, the average jitter, and finally the network throughput.

Many routing protocols have been proposed for mobile ad hoc networks [4]. Each protocol has itsown characteristics and application scenarios. For example, the Optimized Link State Routing Protocol(OLSR) [5] is suitable for high-density MANETs, and the Destination-Sequenced Distance-VectorRouting (DSDV) [6] is more suitable for small-scale MANETs. When the scenario is complex andthe network topology changes rapidly, a single routing protocol cannot meet the requirements ofthe variable node mobility mode, nor can network performance be guaranteed. Accurate perceptionof the surrounding environment and appropriate adjustment strategies of nodes have become thetrends of adaptive routing protocols with the increasing requirements of quality of service (QoS).In response to the above objectives, this paper proposes a solution for choosing routing protocols inhighly dynamic FANETs, namely TARCS (topology change aware-based routing protocol choosingscheme). The TARCS scheme is divided into the following steps:

Firstly, a topology change sensing method is proposed, which can measure the topology changebetween nodes of a high dynamic FANET, and a mobility metric TCD, the topology change degreewith several levels is defined to quantify the perceived results.


Secondly, a heuristic routing strategy of FANETs is proposed using the measurement results ofthe sensing method. The policy compares the measured topology change result with the thresholdreference value, determines the current node movement mode, and appropriately adjusts the routingprotocol used in the network, so that the network routing protocol is more suitable for the changedmobile scenario, and finally the purpose of improving network performance in highly dynamictopology is achieved.

Simulations show that compared with traditional practices, the proper use of TARCS in highlydynamic MANETs in complex scenarios can effectively improve some aspects of network performance.

The content discussed in this article is based on the following assumptions:

• The node motion is simplified to two-dimension movements.• The transmission range of each node is a circular area with the node in the center and a radius of

r. The transmission power of the node is constant within this area.• The number of nodes in the network is constant during the network lifetime.

The remainder of this paper is organized as follows. In Section 2, the TARCS is introduced in detail.In Section 3, the TARCS strategy is validated in the form of simulation experiments. The experimentalresults are discussed in Section 4, and Section 5 contains the conclusion.

2. Related Work

The following paper summarizes related literature from two aspects: topological changeperception and topological change perception processing.

2.1. Topology Change Perception Methods and Processing Methods

The state-of-the-art perceptions of topology changes are limited to mobility aware of nodesand paths. Radhika Ranjan Roy [7] described mobility models and mobility metrics in detail. BaiFan et al. [8] defined several mobile metrics, such as node spatial dependence, node temporaldependence and geographic restrictions to capture mobility characteristics, and proposed theIMPORTANT framework for analyzing the impact of mobility on the performance of routing protocols.Jie Hong et al. [9] explained the relationship among node mobility, network topology changes andnetwork performance such that the high speed of nodes in a MANET did not mean rapid topologychanges, and changes in network topology caused by changes of node speed and direction ultimatelyaffected network performance. A dynamic routing algorithm improved from the dynamic sourcerouting algorithm (DSR [4]) and proposed by Ehssan S. et al. [10] by selecting relatively static nodesaccording to the Doppler frequency shift was applied to the aeronautical ad hoc networks (AANETs)with high-speed nodes. Zheng, Y. et al. [11] proposed the mobility and load aware OLSR (ML-OLSR)protocol, which combined mobile sensing and load sensing with the procedures of selecting more stablenodes as the multi point relay (MPR) points, and lower-load routing to reduce the average end-to-enddelay. The protocol is applicable to the unmanned aerial vehicle (UAV) networks with high-speednodes and unbalanced loads. The protocol proposed by Zhou J.H. et al. [12] was also applicable toAANETs in which the effective time of the largest link was estimated according to the node relativespeed and the Doppler shift of the received signal; meanwhile the traffic load was referenced duringthe route discovery. The mobility pattern aware routing protocol proposed by Hung, C-C. et al. [13]requires geographic assistant location information and relies on IEEE 802.16 base station assistance fora heterogeneous vehicular network (HVN). Athanasios B. et al. [14] proposed a protocol frameworkwith the defined three-layer services of mobility classification service (MCS), strategy selection service(SSS) and routing service (RS) to make nodes with different mobility levels use different routingprotocols. It is suitable for high-speed mobile networks with sparse nodes. Yu, Y. et al. [15] combinedthe DSR protocol with the ant colony optimization algorithm (ACO), judging the stability betweennodes and neighbors according to the received signal power, using the ACO to detect the networkcongestion degree. Nodes with high mobility in the cache were deleted according to the node stability


and the congestion degree. The premise of the protocol proposed by Swidana, A. et al. [16] was that thevelocity of the node had been measured before, the basic idea of which was to prevent high mobilitynodes from participating in route discovery. Khalaf, M. et al. [17] proposed two probability modelsfor speed perception, with which two protocols were proposed by improving the ad-hoc on-demanddistance vector (AODV) [18] routing algorithm. The protocol proposed by Moussaoui, A. et al. [19]judged node stability based on the received signal power, and improved the OLSR protocol by selectingthe more stable neighbor node as the MPR node. Brahmbhatt, S. et al. [20] proposed a protocol to selectthe most powerful node from the received signal as a stable link in light of signal strength-based linkstability estimation to solve the routing failure problem in multipath networks.

2.2. Discussion

The mobility aware methods noted above can be classified into the following categories:

• Neighbor node or link mobility awareness [10–17,19,20].• Neighbor node mobility awareness combined with congestion level perception [11,12,15].• Neighbor node mobility trend prediction [12].

These mobility aware methods can also be divided into location aided [11,13,14,16,17] andnon-location aided [10,12,15,19,20]. The former refers to estimating the neighbor node/path mobilityby means of information from auxiliary devices, such as global positioning system (GPS) receivers,with the advantage of simple and fast calculation and the disadvantage of introducing GPS errors andnew interference. The latter usually measures the position or speed of the neighbor node accordingto the power or the Doppler shift of the received signal. Although no error or other interference isintroduced, the direction and distance of the neighbor node are not accurately determined.

The processing method after mobility awareness can be summarized into the following:

• Selecting stable node/link during route discovery [10–13,17,19].• Excluding or preventing high mobility nodes from participating in route discovery [15,16].• Node mobility classification and decision making [14].

The mobility awareness methods, the processing methods and the proposed TARCS aresummarized in Figure 2.

However, the methods in the above literature are only limited to measuring node mobility orlink stability caused by mobility, without involving accurate definition and accurate measurement oftopology changes.

In this paper, the methods of perception and perceived result processing are different. In termsof the sensing method, the above literature uses only a single parameter (velocity or distance) forperception, but TCD is used here to quantify the topological changes between nodes from multipleaspects, including distance, velocity, direction and neighbors’ number, and to accurately reflect thetopological changes. As to the subsequent processing method, in this paper, after determining thenode’s mobility mode based on the perceived result, the routing protocol is re-selected accordingto the current mobility mode characteristics. Essentially this is a routing protocol optimizationselection strategy.


Electronics 2018, 7, x FOR PEER REVIEW 5 of 20

5

Mobility Aware Methods Processing Methods

Node Location Information

Doppler Frequency Shift

Power of Received Signal

Node mobility classification

Searching more stable routes

Exclude high mobility nodes

Choose right routes

Choose long and stable links

Select a link with a smaller degree of congestion

MUDORML-OLSRNTAR

AMPARPMCS+SSS+RS

APARMDA-AODVSVAP-AODVST-OLSR

…TARCS

QOS requirements

Link lifetime

Load congestion control

Node Speed

Defined measurement terms

Node Stability

Node Mobility Classification

Link Stability Degree, ……

Topology Change Degree

Figure 2. The mobility aware methods and the subsequent processing methods.

However, the methods in the above literature are only limited to measuring node mobility or link stability caused by mobility, without involving accurate definition and accurate measurement of topology changes.

In this paper, the methods of perception and perceived result processing are different. In terms of the sensing method, the above literature uses only a single parameter (velocity or distance) for perception, but TCD is used here to quantify the topological changes between nodes from multiple aspects, including distance, velocity, direction and neighbors’ number, and to accurately reflect the topological changes. As to the subsequent processing method, in this paper, after determining the node's mobility mode based on the perceived result, the routing protocol is re-selected according to the current mobility mode characteristics. Essentially this is a routing protocol optimization selection strategy.

3. A Topology Change Aware Routing Protocol Choosing Scheme (TARCS)

A topology change aware routing protocols choosing scheme (TARCS) is proposed based on the results of the periodic topology change awareness (PTCA) and the adaptive route choosing scheme (ARCS). The purpose of the PTCA is to accurately measure topology changes and determine the movement mode. The purpose of the ARCS is to select an appropriate routing protocol for the current motion mode and to ensure that network performance is affected as little as possible. The periodic topology change perception is based on the topological change degree (TCD), a mobility indicator proposed by the previous study, which could measure the topology changes of nodes and distinguish the Random Waypoint model (RWP [7]), the Reference Point Group Mobility Model (RPGM [7,21]) with different group numbers and the Pursue model [7,21].

Figure 2. The mobility aware methods and the subsequent processing methods.

3. A Topology Change Aware Routing Protocol Choosing Scheme (TARCS)

A topology change aware routing protocols choosing scheme (TARCS) is proposed based on theresults of the periodic topology change awareness (PTCA) and the adaptive route choosing scheme(ARCS). The purpose of the PTCA is to accurately measure topology changes and determine themovement mode. The purpose of the ARCS is to select an appropriate routing protocol for the currentmotion mode and to ensure that network performance is affected as little as possible. The periodictopology change perception is based on the topological change degree (TCD), a mobility indicatorproposed by the previous study, which could measure the topology changes of nodes and distinguishthe Random Waypoint model (RWP [7]), the Reference Point Group Mobility Model (RPGM [7,21])with different group numbers and the Pursue model [7,21].

3.1. Description

The topology change aware routing protocol choosing scheme (TARCS) includes twoprocedures: the periodically topology change aware (PTCA) and the adaptive route choosingscheme (ARCS). Accurate perception of topology changes is a prerequisite for routing protocoladjustment and reselection. The routing protocol selection is based on a preset topology changethreshold/interval value.

Traditional routing protocols either maintain routing paths during node movement (activerouting) or start routing discovery when nodes need to communicate (passive routing). Neitheractive routing nor passive routing pays close attention to the surrounding topology. This is completelysufficient for low-speed nodes, but for high-speed nodes the topology changes so fast that nodescannot respond to the rapidly changing environment and adjust strategy if they lack the knowledgeof the external environment. Obviously, the consequence is that the nodes are not well adapted


to the topology changes, the network performance is severely constrained, and the task cannot besuccessfully completed.

The advantage of TARCS is that:

• It can monitor the topology changes around the nodes in real time.• Nodes will respond to the perceived results in time and adjust to select the appropriate

routing protocol.• It will make the current routing protocol more suitable for node mobility mode and topology

change requirements.

Assume that node A (na) will send information to node B (nb) in a FANET. Figure 3 shows thecomparison of communication mode from na to nb with and without the TARCS. Figure 3a showsthat na communicates with nb directly through a routing protocol in the traditional manner. na firstestablishes a valid path with nb through route discovery, then sends data and maintains the route.Figure 3b shows the communication mode from na to nb after using the TARCS.


6

3.1. Description

The topology change aware routing protocol choosing scheme (TARCS) includes two procedures: the periodically topology change aware (PTCA) and the adaptive route choosing scheme (ARCS). Accurate perception of topology changes is a prerequisite for routing protocol adjustment and reselection. The routing protocol selection is based on a preset topology change threshold/interval value.

Traditional routing protocols either maintain routing paths during node movement (active routing) or start routing discovery when nodes need to communicate (passive routing). Neither active routing nor passive routing pays close attention to the surrounding topology. This is completely sufficient for low-speed nodes, but for high-speed nodes the topology changes so fast that nodes cannot respond to the rapidly changing environment and adjust strategy if they lack the knowledge of the external environment. Obviously, the consequence is that the nodes are not well adapted to the topology changes, the network performance is severely constrained, and the task cannot be successfully completed.

The advantage of TARCS is that:

• It can monitor the topology changes around the nodes in real time. • Nodes will respond to the perceived results in time and adjust to select the appropriate routing

protocol. • It will make the current routing protocol more suitable for node mobility mode and topology

change requirements.

Assume that node A (na) will send information to node B (nb) in a FANET. Figure 3 shows the comparison of communication mode from na to nb with and without the TARCS. Figure 3a shows that na communicates with nb directly through a routing protocol in the traditional manner. na first establishes a valid path with nb through route discovery, then sends data and maintains the route. Figure 3b shows the communication mode from na to nb after using the TARCS.

route protocol

Node A Node B

TARCS

Periodically topology aware

Node A Node BRoute

protocol

Judging mobility mode and re-

choosing route

TCD ≠ 0

TCD = 0

(a) (b)

Figure 3. Communication mode comparison between two nodes without and with the topology change aware routing protocol choosing scheme (TARCS). (a) Traditional communication mode between nodes without the TARCS; (b) communication mode between nodes with the TARCS.

Figure 4 shows the basic workflow of the TARCS. The TARCS is divided into two phases. The left region represents the periodic topology change aware (PTCA), and the right region represents the adaptive routing protocol selection (ARCS).

Figure 3. Communication mode comparison between two nodes without and with the topology changeaware routing protocol choosing scheme (TARCS). (a) Traditional communication mode between nodeswithout the TARCS; (b) communication mode between nodes with the TARCS.


7

Initialize

Get latest TCD

Yes

Judging the motion mode

Routing Re-selecting

Is there any record in neighbor list?

Yes

Yes

No

Update neighbor list: add an item

Update: add new record and calculate new TCD

Does ni received message from neighbor nj of location (Xj,Yj) nad

velocity (Vj)?

Update neighbor list:

Only update time if it is the first

record.Delete neighbor if it is a second time

No

Start topology change aware

TCD <= TCDth ?

No

Waiting for next aware

Data transmisson

Figure 4. Process flow of the TARCS.

3.2. Periodically Topology Change Aware (PTCA)

The purpose of PTCA is to periodically perceive the topology changes of each node and its neighbors within a one-hop transmission range in the network. Since the movement of nodes causes the change of the relative position between the neighboring nodes and in the number of neighboring nodes, thereby causing the network topology to change, the discussion of the topology change begins with the mobility of the node.

Figure 5 shows the topology change caused by the change in distance among nodes. Black dots and red dots represent moving nodes, and dashed circles indicate the effective transmission range of nodes. Figure 5a shows the topology in the initial state. Nodes n1 and n2 are within one hop of n0. Node n3 is within one hop of n1, and n4 is within one hop of n3. Node n0 establishes a route through n1, n3, and n4. From this moment on, n1 starts moving down. Figure 5b shows the network topology after time T. At this time, n1 has moved out of the effective range of n0, and the distance between n0 and n1 increases. If n0 still sends data to n4, it can only pass the path: 𝑛𝑛0 → 𝑛𝑛2 → 𝑛𝑛3 → 𝑛𝑛4. It can be seen that topology changes could be caused by distance changes between nodes.

Figure 4. Process flow of the TARCS.


Figure 4 shows the basic workflow of the TARCS. The TARCS is divided into two phases. The leftregion represents the periodic topology change aware (PTCA), and the right region represents theadaptive routing protocol selection (ARCS).

3.2. Periodically Topology Change Aware (PTCA)

The purpose of PTCA is to periodically perceive the topology changes of each node and itsneighbors within a one-hop transmission range in the network. Since the movement of nodes causesthe change of the relative position between the neighboring nodes and in the number of neighboringnodes, thereby causing the network topology to change, the discussion of the topology change beginswith the mobility of the node.

Figure 5 shows the topology change caused by the change in distance among nodes. Black dotsand red dots represent moving nodes, and dashed circles indicate the effective transmission range ofnodes. Figure 5a shows the topology in the initial state. Nodes n1 and n2 are within one hop of n0.Node n3 is within one hop of n1, and n4 is within one hop of n3. Node n0 establishes a route throughn1, n3, and n4. From this moment on, n1 starts moving down. Figure 5b shows the network topologyafter time T. At this time, n1 has moved out of the effective range of n0, and the distance between n0

and n1 increases. If n0 still sends data to n4, it can only pass the path: n0 → n2 → n3 → n4 . It can beseen that topology changes could be caused by distance changes between nodes.Electronics 2018, 7, x FOR PEER REVIEW 8 of 20

8

n0

n1

n2

n3

n4

n0

n1

n2

n3

n4

(a) (b)

Figure 5. Topology change caused by nodes distance change: (a) topology in the initial state; (b) topology after period T.

Figure 6 shows the topology change caused by changes in the direction of movement between nodes. Figure 6a is the initial state. From this moment on, n0 will move around n1. After time T, the distance between n0 and n1 is still the same. However, there is no effective path between n0 and n4 due to the change of direction between n1 and n0. As is shown in Figure 6b. It can be seen that changes in the direction of movement of nodes can also cause topology changes.

n1

n2

n3

n4

n0

(a)

n2

n3

n4

n1

n0

(b)

Figure 6. Topology change caused by nodes direction change: (a) topology in the initial state; (b) topology after time T.

Figure 7 shows the topology change caused by changes in the relative rate between nodes. Figure 7a is the initial state. All nodes move up, but the nodes move at different rates. The rate of n1 is less than the rate of the others. Figure 7b is the topology after time T. It can be seen that the topology has also changed.

n0 n1

n2

n3

n4

(a)

n0

n1

n2

n3

n4

(b)

Figure 7. Topology change caused by a difference of rate: (a) topology in the initial state; (b) topology after T time.

Figure 5. Topology change caused by nodes distance change: (a) topology in the initial state; (b)topology after period T.

Figure 6 shows the topology change caused by changes in the direction of movement betweennodes. Figure 6a is the initial state. From this moment on, n0 will move around n1. After time T,the distance between n0 and n1 is still the same. However, there is no effective path between n0 and n4

due to the change of direction between n1 and n0. As is shown in Figure 6b. It can be seen that changesin the direction of movement of nodes can also cause topology changes.


8

n0

n1

n2

n3

n4

n0

n1

n2

n3

n4

(a) (b)



n1

n2

n3

n4

n0

(a)

n2

n3

n4

n1

n0

(b)



n0 n1

n2

n3

n4

(a)

n0

n1

n2

n3

n4

(b)


Figure 6. Topology change caused by nodes direction change: (a) topology in the initial state; (b)topology after time T.


Figure 7 shows the topology change caused by changes in the relative rate between nodes.Figure 7a is the initial state. All nodes move up, but the nodes move at different rates. The rate of n1 isless than the rate of the others. Figure 7b is the topology after time T. It can be seen that the topologyhas also changed.


8

n0

n1

n2

n3

n4

n0

n1

n2

n3

n4

(a) (b)



n1

n2

n3

n4

n0

(a)

n2

n3

n4

n1

n0

(b)



n0 n1

n2

n3

n4

(a)

n0

n1

n2

n3

n4

(b)


Figure 7. Topology change caused by a difference of rate: (a) topology in the initial state; (b) topologyafter T time.

Figure 8 illustrates the topology change caused by the number of neighbors. Figure 8a is the initialstate. n1 and n2 are both one-hop neighbors of n0. The initial path of n0 to n4 is n0 → n1 → n3 → n4 .In Figure 8b n1 exits the network for some reason (e.g., hardware failure or energy exhaustion) aftertime T. At this time, the number of neighbors of n0 is reduced resulting in no complete path betweenn0 and n4. Therefore, changes in the number of neighbor nodes can also cause topology changes, butthis condition is not sufficient. For example, although n2 is also a neighbor of n0, the departure of n2

does not affect the path from n0 to n4.


9

Figure 8 illustrates the topology change caused by the number of neighbors. Figure 8a is the initial state. n1 and n2 are both one-hop neighbors of n0. The initial path of n0 to n4 is 𝑛𝑛0 → 𝑛𝑛1 → 𝑛𝑛3 →𝑛𝑛4. In Figure 8b n1 exits the network for some reason (e.g., hardware failure or energy exhaustion) after time T. At this time, the number of neighbors of n0 is reduced resulting in no complete path between n0 and n4. Therefore, changes in the number of neighbor nodes can also cause topology changes, but this condition is not sufficient. For example, although n2 is also a neighbor of n0, the departure of n2 does not affect the path from n0 to n4.

n0

n1

n2

n3

n4

(a)

n0

n2

n3

n4

(b)

Figure 8. Topology change caused by a change in the number of neighbors: (a) topology in the initial state; (b) topology after time T.

From the analysis above, it is concluded that each of these factors, namely, the relative position, motion direction, relative rate, and number of neighbor nodes, may cause a change in the local topology between neighboring nodes. Thus, a new mobility metric, the topology change degree (TCD) was defined by selecting parameters such as node distance, motion direction and movement rate, to measure the topology change between neighbor nodes. The main purpose of TCD is to accurately quantify topology changes between neighboring nodes and quickly identify different motion modes.

Topology change degree between nodes i and j within time T, the 𝑻𝑻𝑻𝑻𝑻𝑻𝒊𝒊,𝒋𝒋 , represents the degree of topology change between the node i and its neighbor node j in the T time period. The detailed definition is shown in Equation (1).

𝑇𝑇𝑇𝑇𝑇𝑇𝑖𝑖,𝑗𝑗(𝑡𝑡, 𝑡𝑡 + 𝑇𝑇) = �𝑤𝑤1 ∙�𝑑𝑑𝑖𝑖,𝑗𝑗𝑡𝑡+𝑇𝑇 − 𝑑𝑑𝑖𝑖,𝑗𝑗𝑡𝑡 �

𝑑𝑑𝑖𝑖,𝑗𝑗𝑡𝑡+ 𝑤𝑤2 ∙

�𝜃𝜃𝑖𝑖,𝑗𝑗𝑡𝑡+𝑇𝑇 − 𝜃𝜃𝑖𝑖,𝑗𝑗𝑡𝑡 �2𝜋𝜋

+ 𝑤𝑤3 ∙�𝑟𝑟𝑒𝑒𝑒𝑒𝑒𝑒𝑖𝑖,𝑗𝑗𝑡𝑡+𝑇𝑇 − 𝑟𝑟𝑒𝑒𝑒𝑒𝑒𝑒𝑖𝑖,𝑗𝑗𝑡𝑡 �

𝑟𝑟𝑒𝑒𝑒𝑒𝑒𝑒𝑖𝑖,𝑗𝑗𝑡𝑡� (1)

where:

• �𝑑𝑑𝑖𝑖,𝑗𝑗𝑡𝑡+𝑇𝑇−𝑑𝑑𝑖𝑖,𝑗𝑗

𝑡𝑡 �

𝑑𝑑𝑖𝑖,𝑗𝑗𝑡𝑡 indicates the change in distance between node i and node j;

• �𝜃𝜃𝑖𝑖,𝑗𝑗𝑡𝑡+𝑇𝑇− 𝜃𝜃𝑖𝑖,𝑗𝑗

𝑡𝑡 �

2𝜋𝜋 indicates the change of direction between node i and node j;

• �𝑟𝑟𝑒𝑒𝑒𝑒𝑒𝑒𝑖𝑖,𝑗𝑗

𝑡𝑡+𝑇𝑇−𝑟𝑟𝑒𝑒𝑒𝑒𝑒𝑒𝑖𝑖,𝑗𝑗𝑡𝑡 �

𝑟𝑟𝑒𝑒𝑒𝑒𝑒𝑒𝑖𝑖,𝑗𝑗𝑡𝑡 indicates the change of relative rate between node i and node j;

• ω is the weight coefficient: here 𝜔𝜔1 = 𝜔𝜔2 = 𝜔𝜔3 = 1.

Average topology change of node i and its neighbors in time T, is define as the 𝑻𝑻𝑻𝑻𝑻𝑻𝒊𝒊,𝒏𝒏𝒏𝒏𝒏𝒏𝒏𝒏(𝒕𝒕, 𝒕𝒕 +𝑻𝑻). Suppose the number of neighbors j of node i is n at time T (1 ≤ j ≤ n), then the average topology change degree of node i and neighbors in time T is defined as the topology change degree average value of each pair of node i and neighbor j, as defined in Equation (2).

𝑇𝑇𝑇𝑇𝑇𝑇𝑖𝑖,𝑛𝑛𝑛𝑛𝑟𝑟𝑛𝑛(𝑡𝑡, 𝑡𝑡 + 𝑇𝑇) =∑ 𝑇𝑇𝑇𝑇𝑇𝑇𝑖𝑖,𝑗𝑗(𝑡𝑡, 𝑡𝑡 + 𝑇𝑇)𝑛𝑛𝑗𝑗=1,𝑗𝑗≠𝑖𝑖

𝑛𝑛 (2)

The whole network topology change in time T, the 𝑻𝑻𝑻𝑻𝑻𝑻𝒏𝒏𝒕𝒕𝒏𝒏𝒏𝒏𝒏𝒏(𝒕𝒕, 𝒕𝒕 + 𝑻𝑻), is defined as the sum of the topological changes of all nodes in the network if the total number of nodes in the network is N (N > n), as shown in Equation (3).

Figure 8. Topology change caused by a change in the number of neighbors: (a) topology in the initialstate; (b) topology after time T.

From the analysis above, it is concluded that each of these factors, namely, the relative position,motion direction, relative rate, and number of neighbor nodes, may cause a change in the localtopology between neighboring nodes. Thus, a new mobility metric, the topology change degree (TCD)was defined by selecting parameters such as node distance, motion direction and movement rate,to measure the topology change between neighbor nodes. The main purpose of TCD is to accuratelyquantify topology changes between neighboring nodes and quickly identify different motion modes.

Topology change degree between nodes i and j within time T, the TCDi,j, represents the degreeof topology change between the node i and its neighbor node j in the T time period. The detaileddefinition is shown in Equation (1).

TCDi,j(t, t + T) =

w1·

∣∣∣dt+Ti,j − dt

i,j

∣∣∣dt

i,j+ w2·

∣∣∣θt+Ti,j − θt

i,j

∣∣∣2π

+ w3·

∣∣∣relvt+Ti,j − relvt

i,j

∣∣∣relvt

i,j

(1)


where:

•∣∣∣dt+T

i,j −dti,j

∣∣∣dt

i,jindicates the change in distance between node i and node j;

•∣∣∣θt+T

i,j −θti,j

∣∣∣2π indicates the change of direction between node i and node j;

•∣∣∣relvt+T

i,j −relvti,j

∣∣∣relvt

i,jindicates the change of relative rate between node i and node j;

• ω is the weight coefficient: here ω1 = ω2 = ω3 = 1.

Average topology change of node i and its neighbors in time T, is define as the TCDi,nbrs(t, t + T).Suppose the number of neighbors j of node i is n at time T (1 ≤ j ≤ n), then the average topologychange degree of node i and neighbors in time T is defined as the topology change degree averagevalue of each pair of node i and neighbor j, as defined in Equation (2).

TCDi,nbrs(t, t + T) =∑n

j=1,j 6=i TCDi,j(t, t + T)

n(2)

The whole network topology change in time T, the TCDntwrk(t, t + T), is defined as the sum ofthe topological changes of all nodes in the network if the total number of nodes in the network is N(N > n), as shown in Equation (3).

TCDntwk(t, t + T) =∑N

i=1 TCDi,nbrs(t, t + T)2

(3)

Experiments have shown that the difference in node mobility affects the network TCD. In addition,TCD has been confirmed to reflect the topology changes around the nodes and the network, and candistinguish several different mobility modes, including RWP and RPGM (with different groups).

The first step of TARCS is to perform periodic topology change perception on each node of theentire network to obtain accurate topology change perception results. The main method of nodemobility mode discrimination of this paper is to compare the topology change perception resultTCDntwk and the topological change threshold reference value TCDTH.

3.3. Adaptive Routing Choosing Scheme (ARCS)

The next step of TARCS is to perform adaptive routing selection, ARCS, after obtaining the resultsof the surrounding topology changes. The ARCS includes node mobility mode discrimination androuting protocol selection. The judgement role is:{

Routing protocol A, i f TCDntwrk ≤ TCDTHRouting protocol B, i f TCDntwrk > TCDTH

(4)

In practical applications, the TCDTH can be pre-calculated with reference to a specific movingmodel. Experiments in Section 4 show the specific calculation process of the threshold value. It isalso possible to set a threshold range for discrimination. If the perceived result TCDntwk is within theallowed interval, the current route protocol is continued to be used, and if the perceived result exceedsthe threshold interval, the routing protocol selection is restarted.

Routing is based on the appropriate mobility model for each routing protocol. Therefore, it isnecessary to calculate and classify possible scenarios, each mobility model with its TCDTH, and thecorresponding routing protocols in advance. After the routing protocol is selected, nodes communicatethrough the newly set routing protocol until the next topology change perception.

Electronics 2019, 8, 274 10 of 19

4. Evaluations and Results

Simulation experiments are designed to evaluate the scheme. It is assumed that all nodes canreceive signals from neighboring nodes within one hop in each direction, the node energy is sufficient,and the transmission distance is constant. The speed of the nodes, the distance between nodes, and thedirection of motion of the nodes are known. Nodes move only in two dimensions.

In order to more clearly reflect the highly dynamic network topology, the Chain scenario [22] isselected to simulate the highly dynamic scene, which combines RWP, RPGM and Pursue model tosimulate nodes to complete a series of tasks such as search/exploration, reconnaissance/patrol andtarget tracking/rescue. The duration is 600 s. From 0 s to 200 s nodes move with the RWP model,then they move with the RPGM (g = 5) model from 200 s to 400 s, and finally nodes move with thePursue model.

The simulation tool is NS-3.25 [23] (Network Simulator 3, Version 3.25). The region is a square of2000 m × 2000 m. The number of nodes is 50 with 50 m/s as the lower speed and the 500 m/s as thehigher speed. The Free space propagation model [23,24] is selected as the propagation mode and theconstant speed propagation delay model is selected as the delay model. The transmission mode isconstant bit rate (CBR) with a bit rate of 16 Kbps and a packet size of 1024 bytes. The MAC (mediaaccess control) protocol is IEEE 802.11g. The alternative routing protocols involved are AODV, OLSR,and DSDV. A detailed description of each mobility model and routing protocol can be found in thereferences. Table 2 lists the experimental parameter items and parameter values.

Table 2. Parameters of simulation experiments.

Parameters Values

Simulation region 2000 m × 2000 mNode number 50

Mobility Model Chain: RWP + RPGM (g = 5) + PursueSimulation Time 602 s

Node speed 50 m/s, 500 m/sRoute protocol OLSR / AODV / DSDV

Propagation model Free space propagation modelDelay model constant speed propagation delay modelData mode CBRPacket size 1024 bytesData rate 16 Kbps

MAC protocol IEEE 802.11bDCFTransmission range 140 m

In the first experiment, the network performance is compared with or without the TARCSscheme. The second experiment is to compare the impact on the network performance with differentTARCS strategies.

4.1. Validation of TARCS

This experiment mainly verified the difference in network performance with or without TARCS.The aware interval is set to 50 s, which means that the topology is perceived 13 times with eachrate. Since the TCDTH are different with different node numbers and node rates, it is necessary topreliminarily set the TCDTH according to the node rate and the specific mobility model.

In this experiment, when node speed is 50 m/s the TCDTH is set as follows:{TCDTH(RWP, RPGM(g = 5)) = 150

TCDTH(RPGM(g = 5), Pursue) = 1000(5)

Electronics 2019, 8, 274 11 of 19

The meaning of TCDTH(RWP, RPGM(g = 5)) is the threshold of the network TCD distinguishingfrom motion with the RWP mode and the RPGM (g = 5) mode. Assuming TCDntwrk is the newlyacquired TCD value of the whole network, the criteria are:

mobility model =

{RWP i f TCDntwrk < TCDTH(RWP, RPGM(g = 5))RPGM(g = 5) i f TCDntwrk > TCDTH(RWP, PGM(g = 5))

(6)

Similarly, TCDTH(RPGM(g = 5), Pursue)) represents the TCD threshold value that candistinguish between motion in RPGM (g = 5) mode and in Pursue mode.

When node speed is 500m/s the TCDTH is configured as follows:{TCDTH(RWP, RPGM(g = 5)) = 1000

TCDTH(RPGM(g = 5), Pursue) = 2000(7)

And the criteria are:

mobility model =

{RPGM(g = 5) i f TCDntwrk < TCDTH(RPGM(g = 5), Pursue)

Pursue i f TCDntwrk > TCDTH(RPGM(g = 5), Pursue)(8)

The calculation method for the threshold value will be introduced in Section 4.3.According to the conclusion of [5], this experiment uses the following routing protocol selection

scheme: the AOPV protocol is used in the RWP scenario, the AODV protocol is used in the RPGMscenario, and the DSDV protocol is used in the Pursue scenario. The TCD perception result at time t isrepresented by TCD(t). The perceptual moment, the actual movement model, the topological changeperception result TCD(t), the judged mobility model, and the selected routing protocol are listed inTable 3.

Table 3. The topological change degree (TCD) perception results and route protocol choosing schemeof Experiment 1.

PerceptualMoment

Actual MobilityModel

Node Speed = 50 m/s Node Speed = 500 m/s RouteProtocolTCD(t) Judged model TCD(t) Judged model

0 RWP 0 RWP 0 RWP AODV50 RWP 2.078 RWP 0 RWP AODV

100 RWP 15.459 RWP 842.592 RWP AODV150 RWP 34.364 RWP 782.536 RWP AODV200 RPGM (g = 5) 177.845 RPGM 589.106 RWP AODV250 RPGM (g = 5) 560.596 RPGM 2513.24 Pursue DSDV300 RPGM (g = 5) 658.254 RPGM 678.321 RWP AODV350 RPGM (g = 5) 493.418 RPGM 2102.88 Pursue DSDV400 Pursue 1963.96 Pursue 4144.1 Pursue DSDV450 Pursue 974.397 RPGM 2042.52 Pursue DSDV500 Pursue 3800.78 Pursue 2312.88 Pursue DSDV550 Pursue 2332.39 Pursue 4482.67 Pursue DSDV600 Pursue 4786.97 Pursue 3666.45 Pursue DSDV

The perceptual results show that among the 26 perceptual moments, the perceived result TCD(t)is less than the threshold value and the determined mobility model does not match the actual modelonly for the speed of 50 m/s at 450 s and the speed of 500 m/s at 250 s, 300 s and 350 s. The rest ofthe perceived results and judgment results are correct. The rest of the perception results are correct,within the threshold range, and the judgment results are consistent with the actual movement model.The perceived efficiency is 84.6%. The above results show that TCD is effective for network topologychange perception, and that the granularity setting and the TCDTH setting are very important forthe judgment.

Electronics 2019, 8, 274 12 of 19

The cause of the error is likely to be the following factors. The threshold setting is not reasonableenough, and the time of each mobility model is not long enough, so the mobility model is not verystable when the nodes move at high speed.

Figure 9 illustrates the comparisons of performance metrics, including the packet delivery ratio,the network throughput, the average end-to-end delay and the average jitter under two differentnode speeds (50 m/s and 500 m/s) with the TARCS (assume that all mobility models are accuratelyidentified)and with other routing protocols.


12

Table 3. The topological change degree (TCD) perception results and route protocol choosing scheme of Experiment 1.

Perceptual moment

Actual mobility

model

Node speed = 50 m/s Node speed = 500 m/s Route

protocol TCD(t) Judged model

TCD(t) Judged model

0 RWP 0 RWP 0 RWP AODV 50 RWP 2.078 RWP 0 RWP AODV 100 RWP 15.459 RWP 842.592 RWP AODV 150 RWP 34.364 RWP 782.536 RWP AODV 200 RPGM (g = 5) 177.845 RPGM 589.106 RWP AODV 250 RPGM (g = 5) 560.596 RPGM 2513.24 Pursue DSDV 300 RPGM (g = 5) 658.254 RPGM 678.321 RWP AODV 350 RPGM (g = 5) 493.418 RPGM 2102.88 Pursue DSDV 400 Pursue 1963.96 Pursue 4144.1 Pursue DSDV 450 Pursue 974.397 RPGM 2042.52 Pursue DSDV 500 Pursue 3800.78 Pursue 2312.88 Pursue DSDV 550 Pursue 2332.39 Pursue 4482.67 Pursue DSDV 600 Pursue 4786.97 Pursue 3666.45 Pursue DSDV

The perceptual results show that among the 26 perceptual moments, the perceived result TCD(t) is less than the threshold value and the determined mobility model does not match the actual model only for the speed of 50 m/s at 450 s and the speed of 500 m/s at 250 s, 300 s and 350 s. The rest of the perceived results and judgment results are correct. The rest of the perception results are correct, within the threshold range, and the judgment results are consistent with the actual movement model. The perceived efficiency is 84.6%. The above results show that TCD is effective for network topology change perception, and that the granularity setting and the TCDTH setting are very important for the judgment.

The cause of the error is likely to be the following factors. The threshold setting is not reasonable enough, and the time of each mobility model is not long enough, so the mobility model is not very stable when the nodes move at high speed.

Figure 9 illustrates the comparisons of performance metrics, including the packet delivery ratio, the network throughput, the average end-to-end delay and the average jitter under two different node speeds (50 m/s and 500 m/s) with the TARCS (assume that all mobility models are accurately identified)and with other routing protocols.

(a)

(b)


13

(c)

(d)

Figure 9. Performance comparisons with and without the TARCS: (a) packet delivery ratio; (b) network throughput; (c) average end-to-end delay; (d) average jitter.

It can be seen from Figure 9a that when node speed is 500 m/s, the packet delivery ratio is 58.9435% only with OLSR, 55.4443% only with DSDV, 74.9448% only with AODV, and rises to 83.3388% with the TARCS.

Figure 9b shows the network throughput comparison with and without TARCS under different node speeds. Taking the speed of 500 m/s as an example, the network throughput is 666.716 Kbps when only OLSR is used, 701.095 Kbps when DSDV is only used, 862.448 Kbps when AODV is only used, and 881.438 Kbps when the TARCS scheme is used. Network throughput is increased by 32.2%, 25.7%, and 2.2%, respectively, compared to using only OLSR, DSDV, and AODV.

Figure 9c shows the average end-to-end delay comparison. Regardless of the rate, the average end-to-end delay is different from the first two performance metrics. Using the TARCS strategy chosen for this experiment, the average end-to-end delay of the network is not the smallest, but between the highest 0.141 s (AODV) and the lowest 0.032 s (DSDV), and is also higher than the OLSR of 0.06 s with the node speed of 500 m/s.

The comparison result of the average jitter is similar to that of the delay, as is shown in Figure 9d. It is between the highest value of 0.149 s (only with AODV) and the lowest value of 0.0383 s (only with DSDV) with the node speed of 500 m/s.

The reason why the advantage of TARCS is not obvious when the speed is low is that the topology change of the network is not obvious when the node speed is low.

The results of first experiment show that:

• In a complex scenario, using the TARCS strategy will have a certain impact on network performance.

• Using the same TARCS strategy at different node speeds has a different impact on network performance.

• In the TARCS strategy, the setting of TCDTH is very important and should be performed in advance.

4.2. Impact Analysis of Different TARCS Strategies on Network Performance

The second experiment is to illustrate different routing strategies can affect different aspects of network performance. This experiment uses three different routing strategies (see Table 4 for

Figure 9. Performance comparisons with and without the TARCS: (a) packet delivery ratio; (b) networkthroughput; (c) average end-to-end delay; (d) average jitter.

It can be seen from Figure 9a that when node speed is 500 m/s, the packet delivery ratio is58.9435% only with OLSR, 55.4443% only with DSDV, 74.9448% only with AODV, and rises to 83.3388%with the TARCS.

Figure 9b shows the network throughput comparison with and without TARCS under differentnode speeds. Taking the speed of 500 m/s as an example, the network throughput is 666.716 Kbpswhen only OLSR is used, 701.095 Kbps when DSDV is only used, 862.448 Kbps when AODV is onlyused, and 881.438 Kbps when the TARCS scheme is used. Network throughput is increased by 32.2%,25.7%, and 2.2%, respectively, compared to using only OLSR, DSDV, and AODV.

Figure 9c shows the average end-to-end delay comparison. Regardless of the rate, the averageend-to-end delay is different from the first two performance metrics. Using the TARCS strategy chosenfor this experiment, the average end-to-end delay of the network is not the smallest, but between thehighest 0.141 s (AODV) and the lowest 0.032 s (DSDV), and is also higher than the OLSR of 0.06 s withthe node speed of 500 m/s.

Electronics 2019, 8, 274 13 of 19

The comparison result of the average jitter is similar to that of the delay, as is shown in Figure 9d.It is between the highest value of 0.149 s (only with AODV) and the lowest value of 0.0383 s (only withDSDV) with the node speed of 500 m/s.

The reason why the advantage of TARCS is not obvious when the speed is low is that the topologychange of the network is not obvious when the node speed is low.

The results of first experiment show that:

• In a complex scenario, using the TARCS strategy will have a certain impact onnetwork performance.

• Using the same TARCS strategy at different node speeds has a different impact onnetwork performance.

• In the TARCS strategy, the setting of TCDTH is very important and should be performedin advance.

4.2. Impact Analysis of Different TARCS Strategies on Network Performance

The second experiment is to illustrate different routing strategies can affect different aspectsof network performance. This experiment uses three different routing strategies (see Table 4for details), and the rest of the parameters are the same as Experiment 1. The first strategy isAODV→AODV→DSDV, i.e., in the RWP model the AODV protocol is used, in the RPGM model theDSDV is used, and in the Pursue model the DSDV protocol is used. It is represented by TARCS1 inTable 3. Similarly, TARCS2 is OLSR→AODV→DSDV and TARCS3 is AODV→DSDV→DSDV.

Table 4. Three different schemes of TARCS.

Mobility Model TARCS1 TARCS2 TARCS3

RWP AODV OLSR AODVRPGM (g = 5) AODV AODV DSDV

Pursue DSDV DSDV DSDV

The network performance under three different types of TARCS strategies is shown in Figure 10.The histogram on the left side of each figure shows the performance for the node speed of 50 m/s, andthe right side shows the performance for the speed of 500 m/s.

It can be seen from Figure 10 that when the node speed is 50 m/s, the packet delivery ratio of thethree strategies has little difference and TARCS1 is slightly dominant (TARCS1: 86.1261%, TARCS2:77.3794%, and TARCS3: 69.3387%). In terms of network throughput, TARCS1 (910.061 Kbps) andTARCS2 (909.755 Kbps) are slightly superior to TARCS3 (766.448 Kbps). In terms of average end-to-enddelay, TARCS3 (0.0495 s) is significantly better than TARCS1 (0.0789 s) and TARCS2 (0.05578 s).Regarding average jitter, TARCS3 (0.0483 s) is better than the other two (TARCS1: 0.0708 s and TARCS2:0.05538 s respectively).

When the node speed is increased to 500 m/s, the packet delivery ratio of TARCS1 is thehighest (sorted in descending order: TARCS1 83.3388%; TARCS3 73.0154%; and TARCS2 69.5115%).The network throughput of TARCS1 is also the highest of the three (sorted in descending order:TARCS1, 881.438 Kbps; TARCS2, 821.2191 Kbps; and TARCS3, 813.157 Kbps). Yet, in terms of averageend-to-end delay, TARCS3 has the lowest value (sorted in ascending order: TARCS3, 0.04968 s,TARCS1, 0.07751 s; and TARCS2, 0.1221 s). The average jitter of TARCS3 is also the lowest (sorted inascending order: TARCS3, 0.04838s; TARCS1, 0.07397 s; and TARCS2, 0.1192 s). None of the networkperformances is effectively improved by TARCS2.

The results of the second experiment indicate that using different TARCS strategies can affectdifferent network performance metrics. The indicators affected is related to the characteristics and theappropriate scenarios of each protocol [5]. The AODV protocol is more suitable for RWP scenarios,while DSDV is more suitable for RPGM (g = 1) scenarios. OLSR sits between these two and is very

Electronics 2019, 8, 274 14 of 19

sensitive to node mobility speed and has higher jitter. The experimental results above also showthis feature. Overall, in order to determine which strategy will get the best performance, in additionto the network performance requirements, the network designer must also deeply understand thecharacteristics of the selected protocol and the suitable scenarios, and make an accurate judgment onthe moving scene of nodes.


14

details), and the rest of the parameters are the same as Experiment 1. The first strategy is AODV→AODV→DSDV, i.e., in the RWP model the AODV protocol is used, in the RPGM model the DSDV is used, and in the Pursue model the DSDV protocol is used. It is represented by TARCS1 in Table 3. Similarly, TARCS2 is OLSR→AODV→DSDV and TARCS3 is AODV→DSDV→DSDV.

Table 4. Three different schemes of TARCS.

Mobility model TARCS1 TARCS2 TARCS3 RWP AODV OLSR AODV

RPGM (g = 5) AODV AODV DSDV Pursue DSDV DSDV DSDV

The network performance under three different types of TARCS strategies is shown in Figure 10. The histogram on the left side of each figure shows the performance for the node speed of 50 m/s, and the right side shows the performance for the speed of 500 m/s.

It can be seen from Figure 10 that when the node speed is 50 m/s, the packet delivery ratio of the three strategies has little difference and TARCS1 is slightly dominant (TARCS1: 86.1261%, TARCS2: 77.3794%, and TARCS3: 69.3387%). In terms of network throughput, TARCS1 (910.061 Kbps) and TARCS2 (909.755 Kbps) are slightly superior to TARCS3 (766.448 Kbps). In terms of average end-to-end delay, TARCS3 (0.0495 s) is significantly better than TARCS1 (0.0789 s) and TARCS2 (0.05578 s). Regarding average jitter, TARCS3 (0.0483 s) is better than the other two (TARCS1: 0.0708 s and TARCS2: 0.05538 s respectively).

When the node speed is increased to 500 m/s, the packet delivery ratio of TARCS1 is the highest (sorted in descending order: TARCS1 83.3388%; TARCS3 73.0154%; and TARCS2 69.5115%). The network throughput of TARCS1 is also the highest of the three (sorted in descending order: TARCS1, 881.438 Kbps; TARCS2, 821.2191 Kbps; and TARCS3, 813.157 Kbps). Yet, in terms of average end-to-end delay, TARCS3 has the lowest value (sorted in ascending order: TARCS3, 0.04968 s, TARCS1, 0.07751 s; and TARCS2, 0.1221 s). The average jitter of TARCS3 is also the lowest (sorted in ascending order: TARCS3, 0.04838s; TARCS1, 0.07397 s; and TARCS2, 0.1192 s). None of the network performances is effectively improved by TARCS2.

(a)

(b)


15

(c)

(d)

Figure 10. Performance comparisons of different TARCS: (a) packet delivery ratio; (b) network throughput; (c) average end-to-end delay; (d) average jitter.

The results of the second experiment indicate that using different TARCS strategies can affect different network performance metrics. The indicators affected is related to the characteristics and the appropriate scenarios of each protocol [5]. The AODV protocol is more suitable for RWP scenarios, while DSDV is more suitable for RPGM (g = 1) scenarios. OLSR sits between these two and is very sensitive to node mobility speed and has higher jitter. The experimental results above also show this feature. Overall, in order to determine which strategy will get the best performance, in addition to the network performance requirements, the network designer must also deeply understand the characteristics of the selected protocol and the suitable scenarios, and make an accurate judgment on the moving scene of nodes.

4.3. Validation of The Topology Change Awareness Method

The experiment in this section is intended to discuss the validity of the topology change awareness method and to explain the calculation of the topological change threshold TCDTH.

We have selected several different forms of node motion, which are RWP, RPGM (g = 50), RPGM (g = 25), RPGM (g = 10), RPGM (g = 5) and Pursue. g represents the group number.

The simulation region is still 2000 m × 2000 m. The duration is 950 s. the number of nodes is 50. The TCDntwrk of different motion patterns is shown in Figure 11.

(a)

(b)

Figure 10. Performance comparisons of different TARCS: (a) packet delivery ratio; (b) networkthroughput; (c) average end-to-end delay; (d) average jitter.


The experiment in this section is intended to discuss the validity of the topology change awarenessmethod and to explain the calculation of the topological change threshold TCDTH.

We have selected several different forms of node motion, which are RWP, RPGM (g = 50), RPGM(g = 25), RPGM (g = 10), RPGM (g = 5) and Pursue. g represents the group number.

The simulation region is still 2000 m × 2000 m. The duration is 950 s. the number of nodes is 50.The TCDntwrk of different motion patterns is shown in Figure 11.

From Figure 11a, it can be seen that:

• The TCDntwrk value is different when nodes move in the different patterns. As the number ofgroups decreases, the value of TCDntwrk gradually increases.

• In the same motion mode, the change of node speed has little effect on the TCDntwrk value.

Electronics 2019, 8, 274 15 of 19


15

(c)

(d)

Figure 10. Performance comparisons of different TARCS: (a) packet delivery ratio; (b) network throughput; (c) average end-to-end delay; (d) average jitter.

The results of the second experiment indicate that using different TARCS strategies can affect different network performance metrics. The indicators affected is related to the characteristics and the appropriate scenarios of each protocol [5]. The AODV protocol is more suitable for RWP scenarios, while DSDV is more suitable for RPGM (g = 1) scenarios. OLSR sits between these two and is very sensitive to node mobility speed and has higher jitter. The experimental results above also show this feature. Overall, in order to determine which strategy will get the best performance, in addition to the network performance requirements, the network designer must also deeply understand the characteristics of the selected protocol and the suitable scenarios, and make an accurate judgment on the moving scene of nodes.


The experiment in this section is intended to discuss the validity of the topology change awareness method and to explain the calculation of the topological change threshold TCDTH.

We have selected several different forms of node motion, which are RWP, RPGM (g = 50), RPGM (g = 25), RPGM (g = 10), RPGM (g = 5) and Pursue. g represents the group number.

The simulation region is still 2000 m × 2000 m. The duration is 950 s. the number of nodes is 50. The TCDntwrk of different motion patterns is shown in Figure 11.

(a)

(b)

Figure 11. Comparisons of topology change degree values with different mobility patterns:(a) TCDntwrk; (b) the average of TCDntwrk (1-RWP, 2-RPGM (g = 50), 3-RPGM (g = 25), 4-RPGM(g = 10), 5-RPGM (g = 5), 6-Pursue).

The explanation for why the TCDntwrk value of RWP model is the smallest and that of Pursuemodel is the largest depends on how the nodes move in these models.

Under the premise of the same area, the same number of nodes and the similar speed, the TCDntwrkof the RWP model is smaller than that of the RPGM model and of the Pursue model, which is causedby the motion pattern of the nodes in the RWP model. When moving in the RWP mode, the nodesare randomly distributed in the active area. Generally, the area of the active area is much larger thanthe transmission range of the nodes. In this case, the motion of the node belongs to the individualmotion, so the number of neighbors is small and not fixed, thus the value of each TCDi,j is small, andthe values of TCDi,nbrs are also small.

When nodes move in groups, such as RPGM (g = 5), every 10 nodes are gathered into one group,and each group moves in an overall manner. The number of neighbors of a node is fixed and mayincrease instantaneously. Because the relative motion of the nodes inner group is weak, and theinter-group relative motion is frequent, when different groups meet and are interlaced, the number ofneighbor nodes may increase instantaneously.

When nodes move in the manner of Pursue, all nodes are gathered into one group. The numberof neighbors of each node is large and fixed, and the node is usually connected with its neighbors,so the TCDi,nbrs value are larger.

Figure 11b shows the average value of each mobility model in Figure 11a. The average valueof TCDntwrk of each mode is more likely to show this difference. A feasible threshold is calculatedby averaging the TCDntwrk average values of the two modes of movement that need to be compared.The threshold values in the above experiments are calculated by this method.

4.4. Analysis of the Influence of Node Density on TCD

In this section, the impact of changes in the number of nodes on TCD results will be analysed.We believe that the number of nodes should be considered together with the range of node activity,

which is reflected in the indicator of node density. Here, the node density in the network is defined asthe number of nodes per unit area within the effective area of the node’s movement in the network.

Node Density =Node Number

Sregion(9)

The revised version gives comparisons of the TCDntwrk when the node density changes. In thecase of group movement, the nodes are relatively concentrated and follow the reference point motion.Taking the RWP model as an example, we analyze the change of TCDntwrk when the number of nodeschanges. There are several cases.

Electronics 2019, 8, 274 16 of 19

Case 1: The area of simulation region is fixed, and the number of nodes changes (the node densitychanges). We take the RWP model as an example. Figure 12a shows the result. It can be seen that in thesame area of region, the value TCDntwrk increases as the density of nodes increases. This phenomenoncan be explained by Equation (2). As the node density increases, the number of node neighborsalso increases.


17

Case 2: The region is proportional to the number of nodes (the node density is constant). Here is also an example of RWP model. In Figure 12b, the node density is fixed to 0.000125. It shows that when the node density is constant, the change in TCD is not so large.

(a)

(b)

Figure 12. Comparisons of TCDntwrk with changed node density and fixed node density: (a) the node density is variable; (b) the node density is fixed.

The value of TCD reflects the degree of topology change between nodes in the network. In fact, it also reflects the degree of aggregation of nodes and the degree of unified action between nodes.

It can be seen from Equations (1) to (3) that:

• For TCDi,j, the factors that affect its value include changes in velocity, direction, and distance between nodes.

• For TCDi,nbrs, in addition to each single TCDi,j value, the number of neighbor nodes within one hop is also one of the influencing factors.

• For TCDntwrk, the number of nodes in the network directly affects the value of TCDntwrk.

4.5. Disccusions

The TARCS routing scheme proposed in this paper is more suitable for complex scenarios with highly dynamic changes in network topology. The advantages are:

1. Node cognition of its surroundings. Moving nodes can sense the topology environment between itself and neighboring nodes accurately and in a timely manner, thereby improving the node's cognitive ability in regards to its environment.

2. Adaptability. Nodes can adjust the routing protocol in real time according to the perceived result, and the adaptability of network is stronger.

3. Scalability. The TARCS adaptive routing mechanism enables a FANET to adapt to changing scenarios. By modifying the topology change degree reference threshold/interval and adding alternate routing protocols, the TARCS can be extended to more scenarios and is compatible with more routing protocols.

In addition, the TARCS routing scheme proposed in this paper also proposes a new processing idea to deal with a highly dynamic MANET, namely, the topology-change-driven protocol selection framework. Compared with the proactive, reactive and hybrid routing protocols, TARCS is more adaptable. The specific description is shown in Table 5.

Figure 12. Comparisons of TCDntwrk with changed node density and fixed node density: (a) the nodedensity is variable; (b) the node density is fixed.

Case 2: The region is proportional to the number of nodes (the node density is constant). Hereis also an example of RWP model. In Figure 12b, the node density is fixed to 0.000125. It shows thatwhen the node density is constant, the change in TCD is not so large.

The value of TCD reflects the degree of topology change between nodes in the network. In fact,it also reflects the degree of aggregation of nodes and the degree of unified action between nodes.

It can be seen from Equations (1) to (3) that:

• For TCDi,j, the factors that affect its value include changes in velocity, direction, and distancebetween nodes.

• For TCDi,nbrs, in addition to each single TCDi,j value, the number of neighbor nodes within onehop is also one of the influencing factors.

• For TCDntwrk, the number of nodes in the network directly affects the value of TCDntwrk.

4.5. Disccusions

The TARCS routing scheme proposed in this paper is more suitable for complex scenarios withhighly dynamic changes in network topology. The advantages are:

1. Node cognition of its surroundings. Moving nodes can sense the topology environment betweenitself and neighboring nodes accurately and in a timely manner, thereby improving the node’scognitive ability in regards to its environment.

2. Adaptability. Nodes can adjust the routing protocol in real time according to the perceived result,and the adaptability of network is stronger.

3. Scalability. The TARCS adaptive routing mechanism enables a FANET to adapt to changingscenarios. By modifying the topology change degree reference threshold/interval and addingalternate routing protocols, the TARCS can be extended to more scenarios and is compatible withmore routing protocols.

In addition, the TARCS routing scheme proposed in this paper also proposes a new processingidea to deal with a highly dynamic MANET, namely, the topology-change-driven protocol selection

Electronics 2019, 8, 274 17 of 19

framework. Compared with the proactive, reactive and hybrid routing protocols, TARCS is moreadaptable. The specific description is shown in Table 5.

Table 5. T Comparisons of topology-change-driven route and other route protocols.

Types of Route Protocols Characteristics Advantages Disadvantages

Proactive routing protocolPeriodically maintain

routing links; suitable forsmall networks.

Short delay; high quality ofservice; long-pass link.

High energy consumption;heavy network load and high

node congestion level.

Reactive routing protocol Establish routing discoverywhen needed.

Low energy consumption; lightrouting load.

The pass-link is notguaranteed. It takes time to

establish a link. Once the linkis interrupted, it needs to be

re-established.

Hybrid routing protocol

Combined with proactiveand reactive routing.

Suitable for hierarchicalnetworks.

Using proactive routing within thecluster, reactive between clusters

Only suitable for hierarchicalnetworks.

Topology-change-drivenprotocol

Reselect the appropriateroute based on topology

changes.

Nodes obtain information of thesurrounding topology changes intime. Multiple routing protocolsare used in combination to suit

complex scenarios.

Dependent on referencethreshold settings and specific

routing strategies.

5. Conclusions

On the basis of analyzing the factors affecting the topology changes between nodes in FANETs,a mobility metric named topology change degree (TCD) is first proposed to describe the topologychanges of highly dynamic FANETs. Meanwhile, a topology change awareness method is applied tomeasure topology changes by periodically sensing the topology changes of FANETs and to distinguishthe mobility modes of nodes. Then a heuristic routing protocol scheme, TARCS, is proposed in thispaper. The TARCS strategy is based on the periodic topology change sensing method, and adaptivelyselects appropriate routing protocols according to the comparison results of the measured topologychange degree and the topology change degree threshold.

Different from the other single routing protocols proposed in other papers, the TARCS proposedin this paper is an adaptive routing strategy. It is based on the topology change perception betweennodes and adaptively selects the routing protocol based on the perceived result. Essentially, it is aflexible routing protocol usage framework. Theoretical analysis and experiments show that:

• The topology change perception method can sense the topology changes of the network anddistinguish several common mobility models of FANET.

• The combined use of routing protocols is more flexible and adaptable than using a single routingprotocol in complex scenarios.

• Proper use of the TARCS scheme in complex scenarios can improve network performance.

Like mobile awareness, topology change perception opens the door to coping with and addressingthe problems of highly dynamic FANETs topology changes. As an adaptive routing protocolframework, TARCS is only one of the applications of topology change perception and it proposes arouting implementation idea in a complex scenario of a FANET network. However, all of the above arejust a preliminary framework, and there are a series of issues that need to be discussed and resolved.For example, how to effectively set the topology change perception interval? How to accuratelycalculate the topology change threshold? How to configure the weight factors among the influencingfactors? How to deal with the problems of changes in the number of nodes? Subsequent work willcontinue to study in depth on the above issues.

Author Contributions: D.Z. is the project administrator. J.H. made the investigation. J.H. and D.Z. proposed themethodology. J.H. did the validation work and finished the original draft writing. D.Z. finished the review &editing writing.

Funding: This research received no external funding.

Electronics 2019, 8, 274 18 of 19

Conflicts of Interest: The authors declare no conflict of interest.

References

1. Guillen-Perez, A.; Cano, M.D. Flying Ad Hoc Network: A New Domain for Network Communications.Sensors 2018, 18, 3571. [CrossRef] [PubMed]

2. Bekmezci, I.; Sahingoz, O.K.; Temel, S. Flying Ad-Hoc Networks (FANETs): A survey. Ad Hoc Netw. 2013, 11,1254–1270. [CrossRef]

3. Oubbati, O.M.; Lakas, A.; Zhou, F.; Günes, M.; Yagoubi, M.B. A survey on position-based routing protocolsfor Flying Ad hoc Networks (FANETs). Veh. Commun. 2017, 10, 29–56. [CrossRef]

4. Sarkar, S.K.; Basavaraju, T.G.; Puttamadappa, C. Routing Protocols. In Ad Hoc Mobile WirelessNetworks-Principles, Protocols, and Applications, 2nd ed.; CRC Press: Boca Raton, FL, USA, 2012; pp. 81–126.

5. Clausen, T.; Jacquet, P. Optimized Link State Routing Protocol (OLSR), RFC 3626 (Experimental). Availableonline: https://www.ietf.org/rfc/rfc3626.txt.pdf (accessed on 9 October 2018).

6. Perkins, C.E.; Bhagwat, P. Highly Dynamic Destination-Sequenced Distance-Vector Routing (DSDV) forMobile Computers. In Proceedings of the conference on Communications architectures, protocols andapplications (SIGCOMM94), London, UK, 31 August–2 September 1994; pp. 234–244.

7. Radhika Ranjan, R. Random Waypoint Mobility, Reference Point Group Mobility. In Handbook of Mobile AdHoc Networks for Mobility Models; Springer: Boston, MA, USA, 2011; pp. 637–670.

8. Fan, B.; Sadagopan, N.; Helmy, A. The Important Framework for Analyzing the Impact of Mobility onPerformance of Routing Protocols for Adhoc Networks. Ad Hoc Netw. 2003, 1, 383–403. [CrossRef]

9. Hong, J.; Zhang, D. Impact Analysis of Node Motion on the performance of FANET routing protocols. InProceedings of the 14th International Conference on Wireless Communications, Networking and MobileComputing (WiCom2018), Chongqing, China, 18–20 September 2018.

10. Sakhaee, E.; Jamalippour, A.; Kato, N. Aeronautical Ad Hoc Networks. In Proceedings of the IEEEWireless Communications and Networking Conference (WCNC 2006), Las Vegas, NV, USA, 3–6 April2006; pp. 246–251. [CrossRef]

11. Zheng, Y.; Wang, Y.; Li, Z.; Dong, L.; Jiang, Y.; Zhang, H. A Mobility and Load aware OLSR routing protocolfor UAV mobile ad-hoc networks. In Proceedings of the 2014 International Conference on Information andCommunications Technologies (ICT2014), Nanjing, China, 15–17 May 2014.

12. Zhou, J.H.; Lei, L.; Liu, W.K.; Tian, J. A simulation analysis of nodes mobility and traffic load aware routingstrategy in aeronautical ad hoc networks. In Proceedings of the 2012 9th International Bhurban Conferenceon Applied Sciences & Technology (IBCAST), Islamabad, Pakistan, 9–12 January 2012; pp. 423–426.

13. Hung, C.-C.; Chan, H.; Wu, E. Mobility Pattern Aware Routing for Heterogeneous Vehicular Networks.In Proceedings of the 2008 IEEE Wireless Communications and Networking Conference (WCNC2008),Las Vegas, NV, USA, 31 March–3 April 2008.

14. Bamis, A.; Boukerche, A.; Chatzigiannakis, I.; Nikoletseas, S. A mobility aware protocol synthesis for efficientrouting in ad hoc mobile networks. Comput. Netw. 2008, 52, 130–154. [CrossRef]

15. Yu, Y.; Ru, L.; Chi, W.; Liu, Y.; Yu, Q.; Fang, K. Ant colony optimization based polymorphism-aware routingalgorithm for ad hoc UAV network. Multimed. Tools Appl. 2016, 75, 14451–14476. [CrossRef]

16. Swidana, A.; Abdelghanya, H.; Saifana, R.; Zilic, Z. Mobility and Direction Aware Ad-hoc on DemandDistance Vector Routing Protocol. Procedia Comput. Sci. 2016, 94, 49–56. [CrossRef]

17. Khalaf, M.; Al-Dubai, Y.; Min, G. New efficient velocity-aware probabilistic route discovery schemes for highmobility Ad hoc networks. J. Comput. Syst. Sci. 2015, 81, 97–109. [CrossRef]

18. Perkins, C.E.; Royer, E.M. Ad-hoc on-demand distance vector routing. In Proceedings of the Second IEEEWorkshop on Mobile Computing Systems and Applications, New Orleans, LA, USA, 25–26 February 1999.

19. Moussaoui, A.; Semchedine, F.; Boukerram, A. A link QoS routing protocol based on link stability for mobilead hoc network. J. Netw. Comput. Appl. 2014, 39, 117–125. [CrossRef]

20. Brahmbhatt, S.; Kulshrestha, A.; Singal, G. SSLSM: Signal Strength Based Link Stability Estimationin MANETs. In Proceedings of the 2015 International Conference on Computational Intelligence andCommunication Networks, Jabalpur, India, 12–14 December 2015.

http://dx.doi.org/10.3390/s18103571

http://www.ncbi.nlm.nih.gov/pubmed/30347892

http://dx.doi.org/10.1016/j.adhoc.2012.12.004

http://dx.doi.org/10.1016/j.vehcom.2017.10.003

https://www.ietf.org/rfc/rfc3626.txt.pdf

http://dx.doi.org/10.1016/S1570-8705(03)00040-4

http://dx.doi.org/10.1109/WCNC.2006.1683472

http://dx.doi.org/10.1016/j.comnet.2007.09.023

http://dx.doi.org/10.1007/s11042-015-3240-y

http://dx.doi.org/10.1016/j.procs.2016.08.011

http://dx.doi.org/10.1016/j.jcss.2014.06.014

http://dx.doi.org/10.1016/j.jnca.2013.05.014

Electronics 2019, 8, 274 19 of 19

21. Hong, X.; Gerla, M.; Pei, G.; Chiang, C.-C. A Group Mobility Model for Ad Hoc Wireless Networks.In Proceedings of the 2nd ACM international workshop on modeling, analysis and simulation of wirelessand mobile systems, Seattle, WA, USA, 20 August 1999; pp. 53–60.

22. Bonnmotion-A Mobility Scenario Generation and Analysis Tool. Available online: http://sys.cs.uos.de/bonnmotion/ (accessed on 10 October 2018).

23. NS-3 Documentation. Available online: https://www.nsnam.org (accessed on 10 October 2018).24. Friis, H.T. A Note on a Simple Transmission Formula. Proc. IRE 1946, 34, 254–256. [CrossRef]

© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open accessarticle distributed under the terms and conditions of the Creative Commons Attribution(CC BY) license (http://creativecommons.org/licenses/by/4.0/).

http://sys.cs.uos.de/bonnmotion/

http://sys.cs.uos.de/bonnmotion/

https://www.nsnam.org

http://dx.doi.org/10.1109/JRPROC.1946.234568

http://creativecommons.org/

http://creativecommons.org/licenses/by/4.0/.

tarcs: a topology change aware-based routing protocol

Documents